基于深度玻尔兹曼模型的红外与可见光图像融合

Infrared and visible image fusion based on deep Boltzmann model

为了克服红外与可见光图像融合时噪声干扰及易产生伪影导致目标轮廓不鲜明、对比度低的缺点,提出一种基于深度模型分割的图像融合方法.首先,采用深度玻尔兹曼机学习红外与可见光的目标和背景轮廓先验,构建轮廓的深度分割模型,通过Split Bregman迭代算法获取最优能量分割后的红外与可见光图像轮廓;然后再使用非下采样轮廓波变换对源图像进行分解,并针对所分割的背景轮廓采用结构相似度的规则进行系数组合;最后进行非下采样轮廓波反变换重构出融合图像.数值试验证明,该算法可以有效获取目标和背景轮廓均清晰的融合图像,融合结果不但具有较高的对比度,还能抑制噪声影响,具有有效性.

In the infrared and visible light image fusion, the noise interference always exists. There is also the disadvantage that image fusion is easy to produce artifacts which cause blurred edge and low contrast. In order to solve these problems, in this study we propose an image fusion method based on deep model segmentation. First of all, deep Bolzmann machine is adopted to learn prior target and background contour and construct a contour deep segmentation model. After the optimal energy segmentation, Split Bregman iteration is used to obtain the infrared and visible image contour. Then non-subsampled contourlet transform is adopted to decompose the source images. The segmented background contour coefficients are fused by the structure similarity rule. Finally, the fused image is reconstructed by the non-subsampled contourlet inverse transform. The experimental results show that this algorithm can effectively obtain fused images with clear target contour and background contour. The fused images also have high contrast and low noise. The results show that it is an effective method of achieving the infrared and visible image fusion.